Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/linville/wirel...
[linux-2.6-block.git] / arch / sparc / kernel / nmi.c
CommitLineData
e5553a6d
DM
1/* Pseudo NMI support on sparc64 systems.
2 *
3 * Copyright (C) 2009 David S. Miller <davem@davemloft.net>
4 *
5 * The NMI watchdog support and infrastructure is based almost
6 * entirely upon the x86 NMI support code.
7 */
8#include <linux/kernel.h>
9#include <linux/param.h>
10#include <linux/init.h>
11#include <linux/percpu.h>
12#include <linux/nmi.h>
13#include <linux/module.h>
14#include <linux/kprobes.h>
15#include <linux/kernel_stat.h>
ffaba674 16#include <linux/reboot.h>
e5553a6d
DM
17#include <linux/slab.h>
18#include <linux/kdebug.h>
19#include <linux/delay.h>
20#include <linux/smp.h>
21
cdd6c482 22#include <asm/perf_event.h>
e5553a6d 23#include <asm/ptrace.h>
e5553a6d
DM
24#include <asm/pcr.h>
25
26/* We don't have a real NMI on sparc64, but we can fake one
27 * up using profiling counter overflow interrupts and interrupt
28 * levels.
29 *
30 * The profile overflow interrupts at level 15, so we use
31 * level 14 as our IRQ off level.
32 */
33
e5553a6d
DM
34static int panic_on_timeout;
35
a8f22264
DM
36/* nmi_active:
37 * >0: the NMI watchdog is active, but can be disabled
38 * <0: the NMI watchdog has not been set up, and cannot be enabled
39 * 0: the NMI watchdog is disabled, but can be enabled
40 */
41atomic_t nmi_active = ATOMIC_INIT(0); /* oprofile uses this */
42EXPORT_SYMBOL(nmi_active);
e5553a6d
DM
43
44static unsigned int nmi_hz = HZ;
a8f22264
DM
45static DEFINE_PER_CPU(short, wd_enabled);
46static int endflag __initdata;
e5553a6d
DM
47
48static DEFINE_PER_CPU(unsigned int, last_irq_sum);
494f6a9e 49static DEFINE_PER_CPU(long, alert_counter);
e5553a6d
DM
50static DEFINE_PER_CPU(int, nmi_touch);
51
52void touch_nmi_watchdog(void)
53{
d89be56b 54 if (atomic_read(&nmi_active)) {
e5553a6d
DM
55 int cpu;
56
57 for_each_present_cpu(cpu) {
58 if (per_cpu(nmi_touch, cpu) != 1)
59 per_cpu(nmi_touch, cpu) = 1;
60 }
61 }
62
63 touch_softlockup_watchdog();
64}
65EXPORT_SYMBOL(touch_nmi_watchdog);
66
67static void die_nmi(const char *str, struct pt_regs *regs, int do_panic)
68{
69 if (notify_die(DIE_NMIWATCHDOG, str, regs, 0,
70 pt_regs_trap_type(regs), SIGINT) == NOTIFY_STOP)
71 return;
72
73 console_verbose();
74 bust_spinlocks(1);
75
76 printk(KERN_EMERG "%s", str);
77 printk(" on CPU%d, ip %08lx, registers:\n",
78 smp_processor_id(), regs->tpc);
79 show_regs(regs);
dc4ff585 80 dump_stack();
e5553a6d
DM
81
82 bust_spinlocks(0);
83
84 if (do_panic || panic_on_oops)
85 panic("Non maskable interrupt");
86
2d0740c4 87 nmi_exit();
e5553a6d
DM
88 local_irq_enable();
89 do_exit(SIGBUS);
90}
91
92notrace __kprobes void perfctr_irq(int irq, struct pt_regs *regs)
93{
94 unsigned int sum, touched = 0;
95 int cpu = smp_processor_id();
96
97 clear_softint(1 << irq);
e5553a6d
DM
98
99 local_cpu_data().__nmi_count++;
100
2d0740c4
DM
101 nmi_enter();
102
e5553a6d
DM
103 if (notify_die(DIE_NMI, "nmi", regs, 0,
104 pt_regs_trap_type(regs), SIGINT) == NOTIFY_STOP)
105 touched = 1;
8183e2b3
DM
106 else
107 pcr_ops->write(PCR_PIC_PRIV);
e5553a6d 108
47a4a0e7 109 sum = kstat_irqs_cpu(0, cpu);
e5553a6d
DM
110 if (__get_cpu_var(nmi_touch)) {
111 __get_cpu_var(nmi_touch) = 0;
112 touched = 1;
113 }
114 if (!touched && __get_cpu_var(last_irq_sum) == sum) {
dd17c8f7
RR
115 __this_cpu_inc(alert_counter);
116 if (__this_cpu_read(alert_counter) == 30 * nmi_hz)
e5553a6d
DM
117 die_nmi("BUG: NMI Watchdog detected LOCKUP",
118 regs, panic_on_timeout);
119 } else {
120 __get_cpu_var(last_irq_sum) = sum;
dd17c8f7 121 __this_cpu_write(alert_counter, 0);
e5553a6d 122 }
a8f22264 123 if (__get_cpu_var(wd_enabled)) {
e5553a6d
DM
124 write_pic(picl_value(nmi_hz));
125 pcr_ops->write(pcr_enable);
126 }
2d0740c4
DM
127
128 nmi_exit();
e5553a6d
DM
129}
130
131static inline unsigned int get_nmi_count(int cpu)
132{
133 return cpu_data(cpu).__nmi_count;
134}
135
e5553a6d
DM
136static __init void nmi_cpu_busy(void *data)
137{
138 local_irq_enable_in_hardirq();
139 while (endflag == 0)
140 mb();
141}
142
143static void report_broken_nmi(int cpu, int *prev_nmi_count)
144{
145 printk(KERN_CONT "\n");
146
147 printk(KERN_WARNING
148 "WARNING: CPU#%d: NMI appears to be stuck (%d->%d)!\n",
149 cpu, prev_nmi_count[cpu], get_nmi_count(cpu));
150
151 printk(KERN_WARNING
152 "Please report this to bugzilla.kernel.org,\n");
153 printk(KERN_WARNING
154 "and attach the output of the 'dmesg' command.\n");
155
a8f22264
DM
156 per_cpu(wd_enabled, cpu) = 0;
157 atomic_dec(&nmi_active);
e5553a6d
DM
158}
159
59abbd1e 160void stop_nmi_watchdog(void *unused)
e5553a6d
DM
161{
162 pcr_ops->write(PCR_PIC_PRIV);
a8f22264
DM
163 __get_cpu_var(wd_enabled) = 0;
164 atomic_dec(&nmi_active);
e5553a6d
DM
165}
166
167static int __init check_nmi_watchdog(void)
168{
169 unsigned int *prev_nmi_count;
170 int cpu, err;
171
a8f22264
DM
172 if (!atomic_read(&nmi_active))
173 return 0;
174
e5553a6d
DM
175 prev_nmi_count = kmalloc(nr_cpu_ids * sizeof(unsigned int), GFP_KERNEL);
176 if (!prev_nmi_count) {
177 err = -ENOMEM;
178 goto error;
179 }
180
181 printk(KERN_INFO "Testing NMI watchdog ... ");
182
183 smp_call_function(nmi_cpu_busy, (void *)&endflag, 0);
184
185 for_each_possible_cpu(cpu)
186 prev_nmi_count[cpu] = get_nmi_count(cpu);
187 local_irq_enable();
188 mdelay((20 * 1000) / nmi_hz); /* wait 20 ticks */
189
190 for_each_online_cpu(cpu) {
a8f22264
DM
191 if (!per_cpu(wd_enabled, cpu))
192 continue;
e5553a6d
DM
193 if (get_nmi_count(cpu) - prev_nmi_count[cpu] <= 5)
194 report_broken_nmi(cpu, prev_nmi_count);
195 }
196 endflag = 1;
a8f22264 197 if (!atomic_read(&nmi_active)) {
e5553a6d 198 kfree(prev_nmi_count);
a8f22264 199 atomic_set(&nmi_active, -1);
e5553a6d
DM
200 err = -ENODEV;
201 goto error;
202 }
203 printk("OK.\n");
204
205 nmi_hz = 1;
206
207 kfree(prev_nmi_count);
208 return 0;
209error:
a8f22264 210 on_each_cpu(stop_nmi_watchdog, NULL, 1);
e5553a6d
DM
211 return err;
212}
213
59abbd1e 214void start_nmi_watchdog(void *unused)
a8f22264
DM
215{
216 __get_cpu_var(wd_enabled) = 1;
217 atomic_inc(&nmi_active);
218
219 pcr_ops->write(PCR_PIC_PRIV);
220 write_pic(picl_value(nmi_hz));
221
222 pcr_ops->write(pcr_enable);
223}
224
225static void nmi_adjust_hz_one(void *unused)
e5553a6d 226{
a8f22264
DM
227 if (!__get_cpu_var(wd_enabled))
228 return;
229
e5553a6d
DM
230 pcr_ops->write(PCR_PIC_PRIV);
231 write_pic(picl_value(nmi_hz));
232
233 pcr_ops->write(pcr_enable);
234}
235
236void nmi_adjust_hz(unsigned int new_hz)
237{
238 nmi_hz = new_hz;
a8f22264 239 on_each_cpu(nmi_adjust_hz_one, NULL, 1);
e5553a6d
DM
240}
241EXPORT_SYMBOL_GPL(nmi_adjust_hz);
242
ffaba674
DM
243static int nmi_shutdown(struct notifier_block *nb, unsigned long cmd, void *p)
244{
a8f22264 245 on_each_cpu(stop_nmi_watchdog, NULL, 1);
ffaba674
DM
246 return 0;
247}
248
249static struct notifier_block nmi_reboot_notifier = {
250 .notifier_call = nmi_shutdown,
251};
252
e5553a6d
DM
253int __init nmi_init(void)
254{
ffaba674
DM
255 int err;
256
a8f22264 257 on_each_cpu(start_nmi_watchdog, NULL, 1);
e5553a6d 258
ffaba674
DM
259 err = check_nmi_watchdog();
260 if (!err) {
261 err = register_reboot_notifier(&nmi_reboot_notifier);
262 if (err) {
a8f22264
DM
263 on_each_cpu(stop_nmi_watchdog, NULL, 1);
264 atomic_set(&nmi_active, -1);
ffaba674
DM
265 }
266 }
59abbd1e 267 if (!err)
cdd6c482 268 init_hw_perf_events();
59abbd1e 269
ffaba674 270 return err;
e5553a6d
DM
271}
272
273static int __init setup_nmi_watchdog(char *str)
274{
275 if (!strncmp(str, "panic", 5))
276 panic_on_timeout = 1;
277
278 return 0;
279}
280__setup("nmi_watchdog=", setup_nmi_watchdog);